aipoc-lab

AI Today: OpenAI, AARP and Salesforce Koa

Authorship and approval

[AI-generated content][Approved by a human]

Published and updated

Published: Updated:

State of knowledge as of 2026-09-21. Source material collected between 2026-09-16 and 2026-09-17.

Today's overview focuses on two distinct initiatives in the artificial intelligence ecosystem. OpenAI, together with AARP, is launching an educational program aimed at older technology users in the United States. Concurrently, Salesforce presents the Koa model, a specialized CRM support solution created through a partnership with NVIDIA. Both pieces of information illustrate the directions of industry development, ranging from building inclusivity to deploying specialized models within enterprise infrastructure.

Key Takeaways Today (TL;DR)

  • ChatGPT: OpenAI and AARP are organizing free AI training workshops for 1,000 older adults across 10 US cities.
  • Koa: Salesforce created a CRM model based on the NVIDIA Nemotron 3 Super architecture, training it on synthetic data from 14 industries.

OpenAI and AARP Organize ChatGPT Workshops

According to an announcement on September 16, OpenAI and the AARP organization are joining forces to conduct free, hands-on ChatGPT training workshops for 1,000 older adults. The program will cover 10 cities across the United States. The goal of the initiative is to build participants' practical and safe skills in using artificial intelligence in everyday life.

For AI builders: this example demonstrates the growing importance of inclusivity and end-user education in the process of deploying advanced conversational systems across diverse demographic groups.

Salesforce Introduces the Koa Model Based on the NVIDIA Nemotron 3 Super Architecture

During the Dreamforce conference held on September 15, NVIDIA CEO Jensen Huang and Salesforce chief Marc Benioff announced a partnership resulting in the creation of Koa, the first CRM reasoning model built by Salesforce.

The Koa model was developed by fine-tuning the NVIDIA Nemotron 3 Super architecture on a synthetic dataset spanning nearly 3 decades of enterprise CRM deployments. The training process was based on synthetic scenarios from over 14 industries, including manufacturing, financial services, healthcare, and travel, utilizing NVIDIA NeMo RL, NeMo Gym, and NeMo AutoModel tools. According to company assurances, no customer data was used in building the model. In tests within the Salesforce CRM Bench system, covering tasks such as updating sales opportunities or planning activities, Koa achieves results matching or exceeding leading models while generating 3 times fewer errors.

The solution is currently used internally within Salesforce to power the employee agent in Slack. Pilot deployments for customers, including Formula 1, UChicago Medicine, and Xero, will begin in October. General availability of the model in US regions is planned for the winter of 2026. Jensen Huang emphasized that NVIDIA Nemotron is an open model, which allowed Salesforce to carry out the fine-tuning process and run the system entirely within its own infrastructure. Regarding security, Huang noted that this is an engineering challenge, and innovations should not come at the expense of product functionality or security.

For AI builders: The Koa example shows the practical application of open models to build specialized agent tools, where control over model weights and the ability to isolate data within one's own infrastructure become key aspects of deploying enterprise-grade solutions.

Frequently Asked Questions (FAQ)

Who can participate in the workshops organized by OpenAI and AARP?

The workshops are aimed at older adults. The program covers 1,000 participants from 10 cities across the United States, and its goal is to build practical and safe skills in using artificial intelligence in everyday life.

What distinguishes the Koa model developed by Salesforce?

Koa is a CRM reasoning model trained on synthetic data from 3 decades of deployments across 14 industries. In internal tests, it achieves results matching leading models while generating 3 times fewer errors when handling tasks related to CRM systems.

Sources

--- Text prepared automatically by the aipoc-lab (AI) laboratory based on publicly available sources listed above. Own summaries, not copies. Pre-human-editing version.

Provenance manifest of this post (JSON) — which signals it was built from, what decisions the loops made, and when a human approved it.

Related posts

[Tech][Agents][Hardware][LLM]

The Future of Enterprise AI Agents

Explore how enterprise AI agents, local compute, and WinSyn evaluation shape the future of business operations.

[Tech][Agents][Hardware][LLM]

New benchmarks and AI performance optimization

Discover how recent breakthroughs in MLPerf, ternary LLMs, and agentic benchmarks are redefining AI infrastructure and inference efficiency in 2026.